Enhancing hazardous material vehicle detection with advanced feature enhancement modules using HMV-YOLO

Area of Science:

• Computer Vision
• Artificial Intelligence
• Road Safety Engineering

Background:

• Road transportation of hazardous chemicals presents significant safety risks.
• Existing vehicle detection systems struggle with small targets and precision.
• Real-time monitoring is crucial for hazardous material transport safety.

Purpose of the Study:

• To develop an enhanced real-time detection system for hazardous material vehicles.
• To improve the accuracy and precision of detecting smaller targets in vehicle monitoring.
• To address the limitations of current detection methods in hazardous material transport.

Main Methods:

• Introduced HMV-YOLO, an enhanced version of the YOLOv7-tiny model.
• Developed two novel modules: CBSG (Convolution, Batch Normalization, SiLU, Global Response Normalization) and G-ELAN.
• CBSG module incorporates GRN to prevent feature collapse and boost neuron activity.
• G-ELAN module enhances feature fusion capabilities.

Main Results:

• The enhanced HMV-YOLO model demonstrated superior performance over the original YOLOv7-tiny.
• Significant improvements were observed across various key evaluation metrics.
• The model shows enhanced capability in detecting smaller targets with higher precision.

Conclusions:

• HMV-YOLO offers a promising advancement for real-time hazardous material vehicle detection.
• The novel CBSG and G-ELAN modules contribute to improved model performance.
• This technology has high potential for practical application in enhancing road safety for hazardous material transport.